The FBI’s Internet Crime Complaint Center recorded more than $3 billion in business email compromise losses last year, in the US alone. Its 2025 report also added a dedicated AI section for the first time, logging $893 million across more than 22,000 complaints involving AI-enabled scams.
The threat isn’t theoretical for an Irish firm. The same techniques are landing in Irish inboxes, and the question for anyone running accounts payable is no longer whether they can spot a suspicious request. It’s whether the process around payments makes fraud difficult regardless of how convincing it looks.
Why accounts payable is the target
AP sits where trust meets time pressure. The team processes invoices, manages supplier details, and executes payments, usually under pressure to keep things moving. For an attacker, that combination is ideal.
Most successful fraud doesn’t involve breaking into systems. It involves impersonating someone the AP team already trusts. A managing partner asking for an urgent payment, a known supplier sending updated bank details, an internal colleague chasing a transfer. IC3’s analysis shows BEC consistently leans on impersonation rather than technical compromise.
AI has made that impersonation dramatically more scalable. Where convincing a finance team once required research, native English, and patience, generative AI now automates all three. By mid-2024, an estimated 40% of BEC phishing emails were already AI-generated.
What AI-enhanced fraud looks like
The modern BEC email is grammatically correct and written in the tone of the executive or supplier being impersonated. It references active projects, current invoice numbers, and upcoming payment runs. For a team processing dozens of routine messages a day, that level of familiarity is exactly what lowers the guard.
Payment redirection is the most common pattern. Attackers intercept a legitimate invoice exchange and quietly alter the destination account. Sometimes they re-issue the same invoice with the IBAN changed by a few digits. Sometimes they send a short note claiming the supplier has updated banking details. The surrounding content reads as legitimate because, in many cases, it is, drawn from real correspondence the attacker has been reading for weeks.
Email isn’t the only channel. AI voice-cloning tools can replicate a person’s voice from a short audio sample. For a team accustomed to a quick phone call confirming an urgent transfer, this removes one of the few verification methods that email security alone can’t address.
Why traditional checks don’t catch it
Security awareness training still matters, and the time invested isn’t wasted. But the signals it used to teach people to watch for, awkward phrasing, mismatched logos, generic greetings, and odd sender addresses, are largely gone from AI-assisted fraud.
When a fraudulent request is indistinguishable from a legitimate one, placing the burden of detection on the AP team puts it in the wrong place. The firms that meaningfully reduce the risk aren’t asking staff to be more suspicious. They’re building verification steps that work regardless of how a message looks.
What actually works
Three pieces of process do most of the heavy lifting.
- Out-of-band verification as standard.
Any request to change supplier bank details, or to approve an urgent payment outside the normal cycle, requires confirmation through a known, independent channel. A phone call to a number already on file. A confirmation in person if the partner is in the office. Not a reply to the same email thread. This step is cultural and procedural, not technical. It works when the team writes it down and follows it consistently. - Layered access controls.
MFA on email, MFA on accounting and banking systems, and limits on who can change supplier records in the first place. If an attacker compromises a supplier’s mailbox, MFA on the receiving end creates the friction that slows or stops a fraudulent change before money moves. - A culture that supports slowing down.
Fraud prevention improves when staff feel safe questioning a request, including from a managing partner. A team member who pauses a payment to verify isn’t being obstructive. They’re doing exactly what good process requires. Building that culture starts with leadership making clear that slowing down on high-value actions is always the right call.
The takeaway
The attackers are using better tools. The process controls that defeat them don’t need to be complicated. They need to be consistent.
If you’d like us to review how your AP process holds up against current BEC patterns, we’re happy to walk through it.